Thread by @JessieSunPsych

I am completely astounded by #ChatGPT's capabilities (chat.openai.com). I've been playing around with it to explore possible use cases for psychological research. Here are just a few: 1) Stimuli generation.

2) Scale development: Generating potential questionnaire items.

3) Providing a "template" for various academic writing tasks. This could be a great teaching tool.

(Of course, ChatGPT can also write the entire research proposal...but 🤫) In this case, interestingly, it got the IV and DV mixed up. Could still be a useful teaching tool to help students learn the research proposal format, instead of working from a blank page.

4) More stimulus generation (this time, focusing on persuasive messages): 🤯🤯🤯

5) Facilitating personality change/goal pursuit: Sometimes when you want to change yourself, it might be difficult to imagine how exactly you would go about doing so. AI can help you come up with strategies! (These could also be used to develop intervention instructions).

6) Understanding the key features of a construct (e.g., honesty/dishonesty), based on how the concept is used in natural language. @CharacterGap you may find this interesting!

You could probably also use the summarizing capabilities to extract themes from participants' open-ended responses! @kkrttr maybe we should try this for comparison.

I might add some more use cases as I think of them, but for now, here are some meta-thoughts: 1) It's made me question how we typically develop stimuli/questionnaire items. Instead of relying on the idiosyncrasies/biases of a small team of researchers/RAs...

ChatGPT uses SO much more information to come up with plausible stimuli. I can imagine this really helping with the issue of stimulus sampling/representativeness, and thus, generalizability.

2) There's an urgent need to develop publication/ethics guidelines for disclosing the use of AI tools in research. Indeed, if you use ChatGPT to generate stimuli, you would have to fully describe the process as part of the Method section.

3) Many RA-intensive tasks (e.g., coding open-ended responses, extracting key themes from open-ended responses) could be largely automated, pending validity evidence that compares ChatGPT with human coding. As ChatGPT currently stores the data for its own use...

We need to think more about the ethical implications of sharing participant data with ChatGPT, and update IRB applications and consent forms accordingly. Perhaps in the future, ChatGPT can be licensed for research/applied use without the data being stored/shared with @OpenAI.

4) I am fairly worried about the prospect of online participants using ChatGPT to generate essay/open-ended text responses. Will we find ways to prevent or detect this, or will we only be able to use writing tasks in the lab going forward?

5) But, despite these and other complexities, I cannot remember the last time I experienced such a sense of amazement and wonder. I'm so glad to be alive in a time of such rapid technological progress, and am excited to find ways to use this as a tool for psychological research!